Table 3.

Log-linear model using demographic variables and baseline utilization phenotype to predict subsequent year primary care telephone encounters and office visits among patients in the test set.

Model predictors	Adjusted R2	AICa
Age-sexb	.166	60,780
Payerc	.128	61,495
Naïve phenotypes (NP)d	.259	57,724
Primary care cluster utilization phenotype (UP)e	.330	55,088
Age-sex and payer	.209	59,450
Age-sex, payer, and NP	.343	54,813
Age-sex, payer, and UP	.394	52,769

^aAIC: Akaike information criterion.

^bAge-sex bins are categorical variables of the combination of male or female with the following age groups: 18-34, 35-49, 50-64, 65-69, 70-84, and 85-115 years.

^cPayers are defined as commercial, Medicare or Medicaid, or other.

^dThe naïve phenotype is a categorical variable that is obtained by summing the total number of health care encounters in the baseline year. These values were rank ordered and divided into 7 percentiles.

^eThe utilization phenotype is a categorical variable encoding 1 of the 7 phenotype clusters created by our algorithm.